Computer vision-based pipeline automatic flue-cured tobacco oven and flue-cured tobacco leaf detection method

Abstract

The present application relates to a pipeline automatic computer vision-based tobacco oven and a tobacco leaf detection method, comprising a plurality of heat preservation boxes, the first of which are connected, a ventilation drying mechanism, a heating mechanism, a closed loop rotary conveying mechanism and a leaf stem clamp are arranged in the heat preservation box; the closed loop rotary conveying mechanism is arranged on the upper end of the heating mechanism, the closed loop rotary conveying mechanism is in a whole back-shaped structure, the closed loop rotary conveying mechanism comprises a track, the track is provided with a leaf stem clamp, and the leaf stem clamp is used for storing tobacco leaves; the ventilation drying mechanism, the heating mechanism and the closed loop rotary conveying mechanism are connected with a control circuit, and the control circuit comprises a detection module and a control module. The present application realizes the drying and sufficient flow of air in the heat preservation box, and makes each side of the tobacco leaf be fully roasted, and the detection module and the control module are arranged and the tobacco leaf detection method is realized, so that the full-automatic high-quality tobacco is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tobacco leaf production, in particular to a pipeline type automatic tobacco leaf curing oven based on computer vision and a method for detecting cured tobacco leaves. BACKGROUND

[0002] The development of tobacco leaf curing houses has gone through a series of developments such as ordinary densification curing houses, changes in curing energy, and automation of curing houses. The three-stage curing process of tobacco is a relatively important process in tobacco processing, which is divided into three stages of yellowing period, color fixing period, and dry muscle period. Due to small-scale planting of tobacco leaves, the same batch of tobacco leaves has different varieties and properties. Therefore, the commonly used three-stage curing process itself has certain defects, resulting in low yield and difficult to guarantee product quality. These defects lead to the phenomenon of green and black tobacco leaves, etc.

[0003] The existing curing house has the problem of uneven curing temperature in different areas due to insufficient air flow and fixed position of tobacco leaves, which eventually causes the phenomenon of green and black tobacco leaves. Secondly, the degree of automation is low, the labor conditions are poor, and the labor intensity is large. The existing tobacco curing house mainly uses an automatic temperature and humidity control system, but the temperature adjustment in the curing process still depends on the subjective judgment of experienced bakers, and then the controller parameters are modified to ensure the curing quality. Since this method does not have a quantitative judgment standard, the curing quality fluctuates greatly, the reliability is not high, it is difficult to guarantee the curing quality, and it also increases the labor cost.

[0004] The quality of cured tobacco products can be judged by the appearance of tobacco leaves. With the development of image recognition technology, the attention mechanism is introduced into the control of tobacco curing equipment, which can greatly improve the automation degree of tobacco curing operation.

[0005] Therefore, there is an urgent need for a tobacco curing oven based on image recognition technology to monitor tobacco leaves by image technology, improve the automation degree of curing, solve the existing problems of the existing tobacco curing house, and improve the quality of cured tobacco. SUMMARY

[0006] The present application provides a pipeline type automatic tobacco leaf curing oven based on computer vision to solve the problem of insufficient air flow and fixed position of tobacco leaves in the existing tobacco curing house, which leads to poor quality of cured tobacco. The air drying mechanism is provided to ensure that the air in the incubator is dry and flows sufficiently. The closed loop rotating conveying mechanism is provided to realize the circulation of tobacco curing, so that each side of the tobacco leaf is fully cured. The detection module and the control module are provided, and the method for detecting cured tobacco leaves is provided to realize fully automatic and high quality curing of tobacco.

[0007] To address the aforementioned issues, the first aspect of this invention proposes a computer vision-based automated tobacco curing oven, comprising an oven, a filling mechanism at one end of the oven, an output mechanism at the other end of the oven, and multiple insulation boxes connected end-to-end. Each insulation box is equipped with a ventilation and drying mechanism, a heating mechanism, a closed-loop rotary conveying mechanism, and a leaf stalk clamp.

[0008] The ventilation and drying mechanism is located on the side wall of the insulated box and is used to dry the air entering the insulated box. The heating mechanism is located at the bottom of the insulated box and is used to heat the air inside the insulated box.

[0009] The closed-loop rotary conveyor mechanism is located at the upper end of the heating mechanism. The closed-loop rotary conveyor mechanism has a U-shaped structure and includes a track. A leaf stalk clamp is provided on the track, and the leaf stalk clamp is used to store tobacco leaves.

[0010] The ventilation and drying mechanism, the heating mechanism, and the closed-loop rotary conveyor mechanism are connected to a control circuit. The control circuit includes a detection module and a control module. The detection module includes a temperature and humidity sensor and an image acquisition device. The control module includes a controller and a communication unit. The controller is connected to a host computer through the communication unit.

[0011] Operating principle: The ventilation and drying mechanism circulates air between the insulation box and the outside air, as well as the air inside the insulation box, achieving air drying. Combined with the control module, it alternates between external and internal circulation. This solves the problem of insufficient airflow in existing tobacco curing barns. The dried air is heated by the heating mechanism for curing the tobacco leaves. The tobacco leaves are placed in the leaf stalk clamps. As the closed-loop rotary conveyor mechanism drives the leaf stalk clamps along a U-shaped path, every side of the tobacco leaf is cured, solving the problem of fixed leaf position.

[0012] A humidity sensor and image acquisition equipment are set up for detection, providing a hardware foundation for image recognition control.

[0013] Furthermore, the ventilation and drying mechanism includes a return air duct, a condenser duct, a ventilation fan assembly, a drainage assembly, and a dry air ventilation duct. The return air duct is located at the top of the insulation box, and the ventilation fan assembly is installed on the return air duct. The ventilation fan assembly includes a connector and a ventilation fan. A condenser duct is installed at the return air inlet of the return air duct, and a drainage assembly and a dry air ventilation duct are installed at the bottom of the condenser duct.

[0014] A ventilation fan is installed on the return air duct to realize the exhaust and intake of air in the insulation box. The intake air is discharged through the drainage component after condensation and sedimentation, and the dry air enters the dry air ventilation duct.

[0015] Further, the heating mechanism comprises a ventilation fan and a heating assembly, the ventilation fan is arranged between the dry air ventilation pipe and the heating assembly;

[0016] The heating assembly and the condensing row are provided with a compressor, the cold end of the compressor is in contact with the condensing row, and the hot end of the compressor is in contact with the heating assembly;

[0017] The compressor, the ventilation fan and the ventilation fan are connected to the power supply through the controller.

[0018] The dry air is heated by the heating assembly and flows in the heat preservation box through the ventilation fan.

[0019] Further, the track is provided with a driving mechanism, the driving mechanism comprises a belt, a rotating shaft and a motor, the rotating shaft is arranged at both ends of the track, the rotating shaft is fixedly connected with the output shaft of the motor, and the belt is sleeved between the two rotating shafts;

[0020] The motor is electrically connected with the controller;

[0021] The driving mechanism is arranged to realize the movement of the petiole clamp.

[0022] Further, the track comprises a No. I track, a No. II track, a No. III track and a No. IV track, the No. I track and the No. II track are connected with arc transition, and the No. III track and the No. IV track are connected with arc transition;

[0023] The No. I track is provided with a turning mechanism at the connection with the No. IV track, and the No. I track is connected with the filling mechanism through the turning mechanism;

[0024] The No. II track is provided with a turning mechanism at the connection with the No. III track, and the No. II track is connected with the output mechanism through the turning mechanism.

[0025] The No. I track, the No. II track, the No. III track and the No. IV track are high on one side and low on the other side, and are connected end to end.

[0026] Further, the turning mechanism comprises a bottom plate, a rotating plate, a baffle and a plurality of driving wheels, the bottom plate is a square plate, and the baffle is arranged on the side of the bottom plate in contact with the No. I track and the No. IV track;

[0027] A circular through hole is formed in the middle of the bottom plate, a rotating plate is arranged in the circular through hole, the rotating plate is a circular plate, the rotating plate is rotatably connected with the bottom plate, a plurality of long strip-shaped through holes are equidistantly formed in the middle of the rotating plate, a driving wheel is arranged in each long strip-shaped through hole, a connecting shaft is arranged between the plurality of driving wheels, the connecting shaft is provided with a driving motor, and the motor shaft of the driving motor is fixed with the connecting shaft;

[0028] The lower end of the rotating plate is provided with a support which is a concave structure, the upper end of the support is fixed with the rotating plate, and the support is arranged outside the connecting shaft and the driving motor, and the driving motor is fixed with the inner side of the support.

[0029] A rotating motor is arranged at the bottom of the support, and the motor shaft of the rotating motor is fixed with the center of the bottom surface of the support.

[0030] The rotating motor and the driving motor are respectively electrically connected with the controller.

[0031] The turning mechanism is arranged to facilitate the turning of the leaf handle clamp, and the turning mechanism ensures the smooth connection of the filling mechanism and the output mechanism with the track.

[0032] Further, the leaf handle clamp comprises a front plate, a rear plate, a vertical ventilation pipe and a lock buckle, the front plate and the rear plate are hinged, the other end of the front plate and the rear plate is fixed through the lock buckle, the side of the front plate and the rear plate is provided with a ventilation opening, a guide vane is arranged at the position corresponding to the ventilation opening, the lower end and the side of the vertical ventilation pipe are open, the vertical ventilation pipe is fixed with the front plate and the rear plate and arranged outside the guide vane.

[0033] The inside of the front plate and the rear plate is provided with a fixed net.

[0034] The leaf handle clamp is arranged at the upper end of the belt.

[0035] The ventilation opening is arranged to facilitate the dry hot air in the heat preservation box to enter the leaf handle clamp to roast the tobacco leaves. The vertical ventilation pipe is arranged to facilitate the collection of the dry hot air flowing from bottom to top.

[0036] Further, the image acquisition device is arranged at both sides of the No. II track.

[0037] The detection module further comprises a photoelectric opposite-shooting switch, the photoelectric opposite-shooting switch is arranged at both ends of the No. I track, the No. II track, the No. III track and the No. IV track respectively, and the photoelectric opposite-shooting switch is used for sensing the leaf handle clamp.

[0038] A plurality of photoelectric opposite-shooting switches are electrically connected with the controller respectively.

[0039] The photoelectric opposite-shooting switch is arranged to facilitate the detection of the position of the leaf handle clamp.

[0040] Further, the controller comprises an MCU chip, and the communication unit comprises one or more combinations of an RFID, a Bluetooth module and a WiFi module.

[0041] The MCU chip is in communication connection with the communication unit through a UART serial port.

[0042] The MCU chip is further connected with a display module and an alarm module, the alarm module comprises a buzzer, a driving circuit and a three-color LED lamp, and the MCU chip is electrically connected with the buzzer and the three-color LED lamp through the driving circuit.

[0043] The MCU chip is connected with the display module in communication through a URAT serial port.

[0044] The MCU chip is connected with the image acquisition device in communication through a USB serial port, and an ADC pin of the MCU chip is connected with the temperature and humidity sensor.

[0045] The embedded device is adopted, voltage level is low, use is safe, and control is convenient.

[0046] The second aspect of the present application proposes a tobacco leaf detection method of a pipeline type automatic tobacco curing furnace based on computer vision, comprising:

[0047] Step 1: constructing a computer vision tobacco maturity recognition model;

[0048] Step 2: constructing a computer vision tobacco recognition classification data set;

[0049] Step 3: stopping the advance of the leaf stalk clamp when the leaf stalk clamp moves to the position of the image acquisition device, and the image acquisition device shoots the front and back images of the leaf blade;

[0050] Step 4: converting the leaf blade into a PNG image;

[0051] Step 5: inputting the image into the computer vision tobacco maturity recognition model to obtain a feature image;

[0052] Step 6: obtaining the main body of the leaf blade according to the feature image anchor frame and the leaf blade smearing area, and archiving the recognized image to form a tobacco leaf archive;

[0053] Step 7: calculating the different maturity marked areas of the front and back leaf blades, and inputting the different area sizes into a classification model to obtain tobacco classification;

[0054] Step 8: when the tobacco classification reaches the target classification requirement, the leaf blade is output.

[0055] Further, the step 2 further comprises: dividing the computer vision tobacco recognition classification data set into a training set and a test set according to 8:2, and inputting the training set into a Kmeans model to train a classification model.

[0056] Further, the computer vision tobacco maturity recognition model comprises 11 layers, wherein:

[0057] The first layer is a convolution layer Conv, which is a single convolution layer, takes a 640*640*3 matrix as input, outputs 64 channels, the convolution kernel size k is 3, and the stride step is 2;

[0058] The 2nd layer is a convolution layer Conv, which is a single convolution layer, takes the output of the 1st layer as input, has an output channel number of 128, a convolution kernel size k of 3, and a stride step of 2;

[0059] The 3rd layer is a channel-to-pixel layer C2f, which is three C2f layers connected by 3 C2f modules, takes the output of the 2nd layer as input, has an output channel number of 128, and has a bottleneck shortcut;

[0060] The 4th layer is a convolution layer Conv, which is a single convolution layer, takes the output of the 3rd layer as input, has an output channel number of 256, a convolution kernel size k of 3, and a stride step of 2;

[0061] The 5th layer is a channel-to-pixel layer C2f, which is six C2f layers connected by 6 C2f modules, takes the output of the 4th layer as input, has an output channel number of 256, and has a bottleneck shortcut;

[0062] The 6th layer is a convolution attention CBAM layer, which is a single CBAM layer, has a channel number of 256, and has a pooling kernel size of 7;

[0063] The 7th layer is a convolution layer Conv, which is a single convolution layer, takes the output of the 6th layer as input, has an output channel number of 512, a convolution kernel size k of 3, and a stride step of 2;

[0064] The 8th layer is a channel-to-pixel layer C2f, which is six C2f layers connected by 6 C2f modules, takes the output of the 7th layer as input, has an output channel number of 512, and has a bottleneck shortcut;

[0065] The 9th layer is a convolution layer Conv, which is a single convolution layer, takes the output of the 8th layer as input, has an output channel number of 1024, a convolution kernel size k of 3, and a stride step of 2;

[0066] The 10th layer is a channel-to-pixel layer C2f, which is three C2f layers connected by 3 C2f modules, takes the output of the 9th layer as input, has an output channel number of 1024, and has a bottleneck shortcut;

[0067] The 11th layer is a spatial pyramid pooling layer SPPF, which takes the output of the 10th layer as input, has a pooling kernel size of 5, and has an output feature map size of 20*20*1024.

[0068] Through the above technical scheme, the application has the following beneficial effects:

[0069] 1. This invention improves the quality of tobacco leaf curing. The invention incorporates a ventilation and drying mechanism to allow exhaust air to flow outwards from the insulated box while injecting outside air into the box. A condenser vent dries the incoming air, reducing moisture content. Temperature and humidity sensors monitor the temperature and humidity within the insulated box, and a controller alternates between internal and external air circulation. This solves the problem of insufficient airflow during curing and improves the quality of the cured tobacco leaves.

[0070] In addition, the present invention is equipped with a closed-loop rotary conveyor mechanism. The closed-loop rotary conveyor mechanism has a U-shaped structure. During the tobacco curing process, the closed-loop rotary conveyor mechanism drives the leaf stalk clamp to make the tobacco leaf move along the U-shaped path. During the movement, the four sides of the leaf stalk clamp will alternate, so that each side of the tobacco leaf is cured for an equal time, thereby solving the problem of the tobacco leaf being fixed in position during the curing process and improving the curing quality of the tobacco leaf.

[0071] 2. This invention uses multiple insulated boxes connected end-to-end. Each insulated box is equipped with a ventilation and drying mechanism, a heating mechanism, a closed-loop conveyor mechanism, and a leaf stalk clamp. Therefore, each insulated box has an independent roasting function. A host computer divides the multiple insulated boxes into multiple roasting processes. After the multiple insulated boxes are connected end-to-end, the tobacco leaves can be roasted at different stages. That is, the finished product roasted in one insulated box enters the next insulated box for subsequent stages of roasting until the tobacco leaves meet the standards, and are then sent out by the output mechanism connected to the last insulated box. This avoids over-roasting and under-roasting, reduces the defect rate, and improves the roasting quality of the tobacco leaves.

[0072] 3. The method of this invention can identify the curing state of flue-cured tobacco leaves, control the closed-loop rotary conveyor mechanism, improve the automation level of the curing process, and improve the quality of tobacco leaf curing. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of the internal structure of the present invention (excluding the side walls of the insulated box).

[0074] Figure 2 This is a schematic diagram of the track structure of the present invention;

[0075] Figure 3 This is a schematic diagram of the steering mechanism structure of the present invention;

[0076] Figure 4 This is a side sectional view of the present invention;

[0077] Figure 5 This is one of the schematic diagrams of the leaf stalk clamp structure of the present invention;

[0078] Figure 6 This is the second schematic diagram of the leaf stalk clamp structure of the present invention;

[0079] Figure 7 This is the electrical schematic diagram of the present invention;

[0080] Figure 8 This is a schematic diagram showing the connection of multiple insulated boxes according to the present invention.

[0081] The attached diagram is labeled as follows: 1 is the insulation box, 2 is the temperature and humidity sensor, 3 is the image acquisition device, 4 is the controller, 5 is the communication unit, 6 is the host computer, 7 is the return air duct, 8 is the condensate drain, 9 is the ventilation fan assembly, 10 is the drainage assembly, 11 is the dry air ventilation duct, 12 is the ventilation fan, 13 is the heating assembly, 14 is the compressor, 15 is track I, 16 is track II, 17 is track III, 18 is track IV, 19 is the belt, 20 is the rotating shaft, 21 is the motor, 22 is the base plate, 23 is the rotating plate, 24 is the baffle, 25 is multiple drive wheels, 26 is the drive motor, 27 is the support, 28 is the rotary motor, 29 is the coupling shaft, 30 is the front plate, 31 is the rear plate, 32 is the vertical ventilation duct, 33 is the latch, 34 is the air guide plate, 35 is the photoelectric beam switch, 37 is the display module, and 38 is the alarm module. Detailed Implementation

[0082] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0083] Example 1

[0084] like Figures 1-7 As shown, a computer vision-based automated tobacco curing oven includes an oven, one end of which is equipped with a filling mechanism and the other end with an output mechanism. The oven includes multiple heat preservation boxes 1 connected end to end. Each heat preservation box 1 is equipped with a ventilation and drying mechanism, a heating mechanism, a closed-loop rotary conveying mechanism, and a blade clamp.

[0085] The ventilation and drying mechanism is located on the side wall of the heat preservation box 1 and is used to dry the air entering the heat preservation box 1. The heating mechanism is located at the bottom of the heat preservation box 1 and is used to heat the air inside the heat preservation box 1.

[0086] The closed-loop rotary conveyor mechanism is located at the upper end of the heating mechanism. The closed-loop rotary conveyor mechanism has a U-shaped structure and includes a track. A leaf stalk clamp is provided on the track, and the leaf stalk clamp is used to store tobacco leaves.

[0087] The ventilation and drying mechanism, the heating mechanism, and the closed-loop rotary conveyor mechanism are connected to a control circuit. The control circuit includes a detection module and a control module. The detection module includes a temperature and humidity sensor 2 and an image acquisition device 3. The control module includes a controller 4 and a communication unit 5. The controller 4 is connected to a host computer 6 through the communication unit 5.

[0088] As Figure 4 shown, preferably, the ventilation drying mechanism includes return air duct 7, condensation drain 8, ventilation fan assembly 9, drainage assembly 10 and dry air duct 11, the return air duct 7 is arranged on the top of the incubator 1, the ventilation fan assembly 9 is arranged on the return air duct 7, the ventilation fan assembly 9 includes a joint and a ventilation fan, the return air outlet of the return air duct 7 is provided with the condensation drain 8, the bottom of the condensation drain 8 is provided with the drainage assembly 10 and the dry air duct 11.

[0089] Preferably, the heating mechanism includes ventilation fan 12 and heating assembly 13, the ventilation fan 12 is arranged between the dry air duct 11 and the heating assembly 13;

[0090] The compressor 14 is arranged between the heating assembly 13 and the condensation drain 8, the cold end of the compressor 14 is in contact with the condensation drain 8, and the hot end of the compressor 14 is in contact with the heating assembly 13;

[0091] The compressor 14, the ventilation fan 12 and the ventilation fan are all connected to the power supply through the controller 4.

[0092] The return air duct 7 is used to collect the exhaust gas in the incubator 1, the ventilation fan assembly 9 is used for ventilation, and the working principle is that the return air duct 7 is communicated with the atmosphere to discharge the exhaust gas, fresh air is input into the condensation drain 8 through the ventilation fan, and in the non-ventilation period, the air in the return air duct 7 is input into the condensation drain 8 using the ventilation fan. In the internal circulation, the condensation drain 8 is used to condense and remove the moisture in the input air, so as to achieve the purpose of reusing the exhaust gas. The drainage assembly 10 is used to collect and discharge the condensed water in the condensation drain 8, and the dry air is introduced into the dry air duct 11. The ventilation fan 12 blows the dry hot air passing through the heating assembly 13 into the incubator 1 to bake the tobacco leaves. As Figure 4 shown. The air path design utilizes the characteristics of hot air rising and cold air sinking, so that the air flow is more smooth. The host computer 6 is used to set the working temperature in the incubator, the temperature of the incubator 1 is detected by using the temperature and humidity sensor 2, and the PID control of the compressor 14 is combined with the main controller, so that the temperature in the incubator 1 is maintained at the working requirement.

[0093] As Figure 5 and 6 shown, preferably, the petiole clamp includes front plate 30, rear plate 31, vertical air duct 32 and lock 33, the front plate 30 and the rear plate 31 are hinged, the other end of the front plate 30 and the rear plate 31 is fixed by the lock 33, the side of the front plate 30 and the rear plate 31 is provided with air vents, the air deflector 34 is arranged at the position corresponding to the air vents, the lower end and the side of the vertical air duct 32 are open, the vertical air duct 32 is fixed with the front plate 30 and the rear plate 31, and is arranged outside the air deflector 34;

[0094] Both the front plate 30 and the rear plate 31 are equipped with fixing mesh inside;

[0095] The petiole clamp is located at the upper end of the belt 10.

[0096] Dry, hot air rises under the action of the ventilation fan 12, enters the vertical ventilation pipe 32, and enters the leaf stalk clamp through the ventilation opening under the action of the air guide plate 34, baking the tobacco leaves inside the leaf stalk clamp.

[0097] Example 2

[0098] The closed-loop rotary conveyor mechanism is optimized based on the above embodiments.

[0099] like Figures 1-4 As shown, preferably, the track is provided with a drive mechanism, which includes a belt 19, a rotating shaft 20 and a motor 21. The rotating shaft 20 is disposed at both ends of the track, and the rotating shaft 20 is fixedly connected to the output shaft of the motor 21. The belt 10 is sleeved between the two rotating shafts 20.

[0100] The motor 21 is electrically connected to the controller 4.

[0101] Preferably, the track includes track I 15, track II 16, track III 17 and track IV 18, with the connection between track I 15 and track II 16 having a rounded transition, and the connection between track III 17 and track IV 18 having a rounded transition.

[0102] A steering mechanism is installed at the connection between track I 15 and track IV 18. Track I 15 is connected to the loading mechanism through the steering mechanism.

[0103] A steering mechanism is installed at the connection between track II 16 and track III 17. Track II 16 is connected to the output mechanism through the steering mechanism.

[0104] Preferably, the steering mechanism includes a base plate 22, a rotating plate 23, a baffle 24 and a plurality of drive wheels 25. The base plate 22 is a square plate, and the baffle 24 is provided on the side of the base plate 22 that contacts track I 15 and track IV 18.

[0105] A circular through hole is provided in the middle of the base plate 22. A rotating plate 23 is provided in the circular through hole. The rotating plate 23 is a circular plate and is rotatably connected to the base plate 22. Multiple elongated through holes are provided at equal intervals in the middle of the rotating plate 23. A drive wheel 25 is provided in each of the elongated through holes. A connecting shaft 29 is provided between the multiple drive wheels 25. A drive motor 26 is provided in the connecting shaft 29. The motor shaft of the drive motor 26 is fixed to the connecting shaft 29.

[0106] The lower end of the rotating plate 23 is provided with a support 27, which is a concave structure. The upper end of the support 27 is fixed to the rotating plate 23. The support 27 is arranged outside the connecting shaft 29 and the driving motor 26. The driving motor 26 is fixed to the inner side of the support 27.

[0107] A rotating motor 28 is arranged at the bottom of the support 27. The motor shaft of the rotating motor 28 is fixed to the center of the bottom surface of the support 27.

[0108] The rotating motor 28 and the driving motor 26 are respectively electrically connected to the controller 4.

[0109] When the leaf handle is clamped at the contact position of the No. I track 15 and the filling mechanism, in order to realize that the leaf handle clamp on the filling mechanism can smoothly enter the No. I track 15 and will not affect the leaf handle clamp on the No. IV track 18 to enter the No. I track 15 during baking, a turning mechanism is arranged between the No. II track 16 and the output mechanism, and the reason for arranging the turning mechanism is the same as above.

[0110] During operation, the controller controls the motor 21 to be powered on. The belt 19 is driven under the action of the rotating shaft 20. The leaf handle clamp sequentially passes through the No. I track 15 and the No. II track 16 and enters the No. III track 17 through the turning mechanism. The No. III track 17 and the No. IV track 18 are transmitted through the belt 19, and the No. IV track 18 enters the No. I track 15 in circulation through the turning mechanism. During the transmission process, the four side surfaces of the leaf handle clamp sequentially contact with the dry hot air, thereby ensuring the uniformity of baking.

[0111] At the junction of the No. I track 15 and the No. IV track 18, and the junction of the No. II track 16 and the No. III track 17, the controller 4 controls the driving motor 26 to be powered on. The driving motor 26 drives the driving wheel 25 to make the driving wheel 25 continue to drive the leaf handle clamp to enter the next track.

[0112] When the leaf handle clamp enters the No. I track 15 from the filling mechanism or leaves the No. II track 16 to enter the output mechanism, the controller makes the rotating motor 28 rotate. The support 27 drives the rotating plate 23 to rotate by a fixed angle. At this moment, the movement direction of the driving wheel 25 changes, which promotes the leaf handle clamp to move towards the filling mechanism or the output mechanism, thereby realizing the reversing of the leaf handle clamp.

[0113] Embodiment 3

[0114] Preferably, the image acquisition device 3 is arranged on both sides of the No. II track 16.

[0115] The detection module further comprises photoelectric opposite-acting switches 35, which are respectively arranged at both ends of the No. I track 15, the No. II track 16, the No. III track 17 and the No. IV track 18. The photoelectric opposite-acting switches 35 are used for sensing the leaf handle clamp.

[0116] A plurality of photoelectric opposite-acting switches 35 are respectively electrically connected to the controller 4.

[0117] Preferably, the controller 4 comprises an MCU chip, and the communication unit 5 comprises one or a combination of RFID, Bluetooth module and WiFi module;

[0118] The MCU chip is in communication connection with the communication unit 5 through a URAT serial port;

[0119] The MCU chip is further connected with a display module 37 and an alarm module 38, the alarm module 38 comprising a buzzer, a driving circuit and a three-color LED lamp, and the MCU chip is electrically connected with the buzzer and the three-color LED lamp through the driving circuit;

[0120] The MCU chip is in communication connection with the display module 37 through a URAT serial port;

[0121] The MCU chip is in communication connection with the image acquisition device 3 through a USB serial port, and an ADC pin of the MCU chip is connected with the temperature and humidity sensor 2.

[0122] In the embodiment, the MCU chip adopts an STM32 single-chip microcomputer, the communication unit 5 adopts a WiFi module, and a driving circuit is arranged between the MCU chip and the compressor 14, the ventilation fan 12 and the air exchange fan, the motor 21, the driving motor 26 and the rotating motor 28, and the power or rotating angle of the above-mentioned electrical equipment is adjusted through a PWM control mode.

[0123] The host computer 6 is configured to have a Windows 10 operating system, a Python 3.9 programming language, a 32G memory, a PyTorch framework technology, a CUDA 11.6 acceleration environment, a NVIDIA GeForce RTX 3090 GPU and a 24G video memory.

[0124] The image acquisition device 3 adopts a 1080p panoramic monitoring camera and acquires images through a USB-CAM mode.

[0125] The MCU chip can determine the position of the petiole clamp according to the signal emitted by the photoelectric reflection switch 35.

[0126] As shown in FIGS. Figures 1-4 and 7, the following method is used to control the closed-loop rotating conveying mechanism:

[0127] Step 1: build a computer vision tobacco maturity recognition model;

[0128] Step 2: build a computer vision tobacco recognition classification data set;

[0129] Step 3: stop the petiole clamp when it moves to the position of the image acquisition device 3, and the image acquisition device 3 shoots the images of the front and back of the leaf;

[0130] Step 4: Convert the leaf into a PNG image with a format of 640*640*3;

[0131] Step 5: Input the image into the computer vision tobacco maturity recognition model to obtain a feature image with a size of 20*20*1024;

[0132] Step 6: Obtain the leaf body according to the feature image anchor frame and the leaf smearing area, and archive the recognized image to form a flue-cured tobacco leaf archive;

[0133] Step 7: Calculate the different maturity labeling areas of the front and back of the leaf, and input the different area sizes into the classification model to obtain the classification of flue-cured tobacco;

[0134] Step 8: When the classification of the tobacco leaf reaches the target classification requirement, the leaf is output.

[0135] In this embodiment, the computer vision tobacco recognition classification data set described in step 2 contains different maturity flue-cured tobacco leaf images, and the acquisition method is as follows:

[0136] Put the different maturity leaf samples into the petiole clamp, put the petiole clamp into the flue-cured tobacco oven, start the sample collection function of the flue-cured tobacco oven control program, input a leaf clamp into the flue-cured tobacco oven, and after reaching the image sensor area, the image acquisition device 3 collects the images of the two sides of the tobacco leaf, saves them to the controller storage space after numbering, and then outputs the leaf clamp after collection. Repeat the collection until all the leaf clamps in the input track are collected, package the images in the storage into a data set to be labeled.

[0137] In order to improve the detection accuracy, the data set to be labeled is labeled to form a computer vision tobacco recognition classification data set. The labeling method is as follows:

[0138] Define different maturity types and assign a color, select a labeling color for the tobacco body, open an image, clear all existing labeling data, use the anchor frame method to frame the main area of the tobacco leaf, use the smearing method to smear the tobacco body in the image with the labeling color, use the smearing method to smear the areas in the image that meet the characteristics of different maturity with the corresponding color, convert the areas to labeling data format according to the smearing areas, save them as labeling files, repeat the above steps until all images are labeled, convert all images to PNG format images with a size of 640*640*3, package all labeling files and images into a recognition data set. Count the labeling area of each image in the data set, classify the leaf maturity, and form a classification data set.

[0139] Further, the computer vision tobacco leaf recognition classification dataset needs to use the dataset to train the model parameters of each layer to form a usable computer vision tobacco leaf recognition classification dataset. Specifically: the leaf body is obtained by excluding interference according to the anchor frame cutout of the recognition dataset image, and the leaf body range is further refined according to the leaf annotation region cutout;

[0140] The leaf annotation data outside the leaf annotation is removed, the recognition dataset is divided into a training set and a test set according to 8:2, the training set is input into the model for training, the test set is used to detect the accuracy, recall rate and average deviation of the model, the training is repeated until the model converges, the accuracy, recall rate and average deviation value tend to be stable, and the model with the optimal evaluation index is selected as the final maturity recognition model;

[0141] Finally, the computer vision tobacco leaf recognition classification dataset is divided into a training set and a test set according to 8:2, and the training set is input into the Kmeans model to train a classification model.

[0142] Preferably, the computer vision tobacco leaf maturity recognition model includes 11 layers, wherein:

[0143] The first layer is a convolution layer Conv, which is a single convolution layer, takes a 640*640*3 matrix as input, outputs 64 channels, the convolution kernel size k is 3, and the stride step is 2;

[0144] The second layer is a convolution layer Conv, which is a single convolution layer, takes the output of the first layer as input, outputs 128 channels, the convolution kernel size k is 3, and the stride step is 2;

[0145] The third layer is a channel-to-pixel layer C2f, which is three C2f layers connected by three C2f modules, takes the output of the second layer as input, outputs 128 channels, and the Bottleneck has a shortcut;

[0146] The fourth layer is a convolution layer Conv, which is a single convolution layer, takes the output of the third layer as input, outputs 256 channels, the convolution kernel size k is 3, and the stride step is 2;

[0147] The fifth layer is a channel-to-pixel layer C2f, which is six C2f layers connected by six C2f modules, takes the output of the fourth layer as input, outputs 256 channels, and the Bottleneck has a shortcut;

[0148] The sixth layer is a convolution attention CBAM layer, which is a single CBAM layer, has 256 channels, and the pooling kernel size is 7;

[0149] The seventh layer is a convolution layer Conv, which is a single convolution layer, takes the output of the sixth layer as input, outputs 512 channels, the convolution kernel size k is 3, and the stride step is 2;

[0150] The 8th layer is a channel-to-pixel layer C2f, which consists of six C2f modules connected together. It takes the output of the 7th layer as input and has 512 output channels. The Bottleneck has a shortcut.

[0151] The 9th layer is a convolutional layer Conv, which is a single-layer convolutional layer. It takes the output of the 8th layer as input, has 1024 output channels, a kernel size k of 3, and a stride of 2.

[0152] The 10th layer is a channel-to-pixel C2f layer, which is a three-layer C2f layer with three C2f modules connected together. It takes the output of the 9th layer as input and has 1024 output channels. The Bottleneck has a shortcut.

[0153] The 11th layer is a Spatial Pyramid Pooling (SPPF) layer, which takes the output of the 10th layer as input, has a pooling kernel size of 5, and outputs a feature map size of 20*20*1024.

[0154] Example 4

[0155] The host computer 6 runs the oven control program. Typically, one host computer 6 controls multiple oven production lines, such as... Figure 8 As shown, in this embodiment, an oven is provided with four warming boxes 1, which are connected end to end (the turning mechanism positions at track II 16 and track III 17 of the previous warming box 1 are connected to the turning mechanism positions at track I 15 and track IV 18 of the next warming box 1 outside the warming box 1). The first warming box 1 is connected to the filling mechanism, and the fourth warming box 1 is connected to the output mechanism.

[0156] The four insulated boxes are divided into a first-stage tobacco curing barn, a second-stage tobacco curing barn, a third-stage tobacco curing barn, and a finished product storage room.

[0157] The operating parameters of a tobacco curing barn are controlled by a host computer, which is set to 38 degrees Celsius and 80% RH. The barn will shut down when the curing conditions are met.

[0158] The second-stage tobacco curing barn is controlled by the host computer 6, with the parameters set at 42 degrees Celsius and 70% RH, and it will exit when the semi-cured condition is reached.

[0159] The three-stage tobacco curing barn is controlled by a host computer with operating parameters set at 54 degrees Celsius and 30% RH, and will exit when the conditions for full curing are met.

[0160] The finished tobacco curing barn is controlled by the host computer 6, with the parameters set at 68 degrees Celsius, 20% RH, and the curing time is 3 hours before exiting.

[0161] Four insulated boxes are interconnected to process different processes separately, reducing the defect rate and improving the quality of baking.

[0162] The preferred embodiments of the present application have been described in detail above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, and various modifications or replacements of the technical solutions of the present application, which are equivalent or equivalent to the above-described embodiments, fall within the scope of the present application without departing from the spirit of the present application, i.e., the scope of disclosure.

Claims

1. A computer vision-based pipeline automatic tobacco curing furnace, comprising a curing oven, one end of the curing oven is provided with a loading mechanism, and the other end of the curing oven is provided with an output mechanism, characterized in that, The oven comprises a plurality of heat preservation boxes (1), the heat preservation boxes (1) are connected end to end, a ventilation and drying mechanism, a heating mechanism, a closed loop rotating conveying mechanism and a leaf stalk clamp are arranged in the heat preservation box (1); The ventilation and drying mechanism is arranged on the side wall of the heat preservation box (1) and is used for drying the air entering the heat preservation box (1); the heating mechanism is arranged on the bottom of the heat preservation box (1) and is used for heating the air in the heat preservation box (1); The closed loop rotating conveying mechanism is arranged on the upper end of the heating mechanism and has a whole U-shaped structure, the closed loop rotating conveying mechanism comprises a track, the track is provided with the leaf stalk clamp, and the leaf stalk clamp is used for storing tobacco leaves; The ventilation and drying mechanism, the heating mechanism and the closed loop rotating conveying mechanism are connected with a control circuit, the control circuit comprises a detection module and a control module, the detection module comprises a temperature and humidity sensor (2) and an image acquisition device (3), the control module comprises a controller (4) and a communication unit (5), and the controller (4) is connected with an upper computer (6) through the communication unit (5); The ventilation and drying mechanism comprises a return air pipe (7), a condensation row (8), a ventilation fan assembly (9), a drainage assembly (10) and a dry air ventilation pipe (11), the return air pipe (7) is arranged on the top of the heat preservation box (1), the ventilation fan assembly (9) is arranged on the return air pipe (7), the ventilation fan assembly (9) comprises a joint and a ventilation fan, the return air port of the return air pipe (7) is provided with the condensation row (8), the bottom of the condensation row (8) is provided with the drainage assembly (10) and the dry air ventilation pipe (11); The heating mechanism comprises a ventilation fan (12) and a heating assembly (13), and the ventilation fan (12) is arranged between the dry air ventilation pipe (11) and the heating assembly (13); The heating assembly (13) and the condensation row (8) are provided with a compressor (14), the cold end of the compressor (14) is in contact with the condensation row (8), and the hot end of the compressor (14) is in contact with the heating assembly (13); The compressor (14), the ventilation fan (12) and the ventilation fan are connected with a power supply through the controller (4); The track is provided with a driving mechanism, the driving mechanism comprises a belt (19), a rotating shaft (20) and a motor (21), the rotating shaft (20) is arranged at two ends of the track, the rotating shaft (20) is fixedly connected with the output shaft of the motor (21), and the belt (10) is sleeved between the two rotating shafts (20); The motor (21) is electrically connected with the controller (4); The track comprises an I-shaped track (15), a II-shaped track (16), a III-shaped track (17) and a IV-shaped track (18), the I-shaped track (15) and the II-shaped track (16) are connected through a circular arc, and the III-shaped track (17) and the IV-shaped track (18) are connected through a circular arc; A turning mechanism is arranged at the connection position of the I-shaped track (15) and the IV-shaped track (18), and the I-shaped track (15) is connected with a filling mechanism through the turning mechanism. The turning mechanism is arranged at the connection between the second track (16) and the third track (17), and the second track (16) is connected to the output mechanism through the turning mechanism; The image acquisition device (3) is arranged on both sides of the second track (16); The detection module further comprises photoelectric opposite-acting switches (35), which are arranged at both ends of the first track (15), the second track (16), the third track (17) and the fourth track (18), respectively, and are used for sensing the petiole clamp. The plurality of photoelectric opposite-acting switches (35) are electrically connected with the controller (4).

2. The computer vision based pipelined automatic flue-cured tobacco oven of claim 1, wherein, The turning mechanism comprises a bottom plate (22), a rotating plate (23), a baffle (24) and a plurality of drive wheels (25), the bottom plate (22) is a square plate, and the baffle (24) is arranged on the side of the bottom plate (22) in contact with the first track (15) and the fourth track (18); A circular through hole is formed in the middle of the bottom plate (22), the rotating plate (23) is arranged in the circular through hole, the rotating plate (23) is a circular plate, the rotating plate (23) is rotationally connected with the bottom plate (22), a plurality of long strip-shaped through holes are equidistantly formed in the middle of the rotating plate (23), the drive wheels (25) are arranged in the long strip-shaped through holes, respectively, a connecting shaft (29) is arranged between the plurality of drive wheels (25), the connecting shaft (29) is provided with a drive motor (26), and the motor shaft of the drive motor (26) is fixed with the connecting shaft (29); A support (27) is arranged at the lower end of the rotating plate (23), the support (27) is a concave structure, the upper end of the support (27) is fixed with the rotating plate (23), the support (27) is arranged outside the connecting shaft (29) and the drive motor (26), and the drive motor (26) is fixed with the inner side surface of the support (27); A rotary motor (28) is arranged at the bottom of the support (27), and the motor shaft of the rotary motor (28) is fixed with the center of the bottom surface of the support (27); The rotary motor (28) and the drive motor (26) are electrically connected with the controller (4), respectively.

3. The computer vision based pipelined automatic flue-cured tobacco oven of claim 1, wherein, The petiole clamp comprises a front plate (30), a rear plate (31), a vertical ventilation pipe (32) and a lock buckle (33), the front plate (30) and the rear plate (31) are hingedly connected, the other ends of the front plate (30) and the rear plate (31) are fixed through the lock buckle (33), ventilation openings are formed in the side portions of the front plate (30) and the rear plate (31), respectively, air deflectors (34) are arranged at positions corresponding to the ventilation openings, the lower end and the side portion of the vertical ventilation pipe (32) are open, the vertical ventilation pipe (32) is fixed with the front plate (30) and the rear plate (31) and is arranged outside the air deflector (34); The inside of the front plate (30) and the rear plate (31) is provided with a fixing net; The petiole clamp is arranged on the upper end of the belt (10).

4. The method for detecting tobacco leaves in a computer vision-based automatic pipeline tobacco oven, applied to the computer vision-based automatic pipeline tobacco oven according to any one of claims 1-3, characterized in that, The method comprises the following steps: Step 1: constructing a computer vision tobacco maturity recognition model; Step 2: Construct a computer vision tobacco leaf recognition classification dataset containing images of different ripeness levels. The acquisition method is as follows: load different ripeness leaf samples into leaf stalk clamps, place the leaf stalk clamps into the tobacco oven, start the tobacco oven control program sample collection function, control the program to input a leaf stalk clamp into the tobacco oven, when the leaf stalk clamp reaches the image acquisition device area, the image acquisition device (3) acquires images of both sides of the tobacco leaf, numbers the images and saves them to the controller storage space, after acquisition, the leaf stalk clamp is output, the acquisition is repeated until all leaf stalk clamps in the input track are collected, and the images in the storage are packaged to form a dataset to be labeled; After labeling the dataset to be labeled, a computer vision tobacco leaf recognition classification dataset is formed. The labeling method is as follows: define different ripeness types and assign a color to each type, select a labeling color for the tobacco leaf body, open an image, remove all existing labeling data, use the anchor box method to frame the main area of the tobacco leaf, use the smearing method to smear the tobacco leaf body in the image with the labeling color, use the smearing method to smear the areas in the image that meet the characteristics of different ripeness levels with the corresponding color, convert the smearing areas to labeling data format, save them as labeling files, and repeat the above steps until all images are labeled; Step 3: When the leaf stalk clamp moves to the image acquisition device (3) position, it stops advancing, and the image acquisition device (3) takes pictures of the front and back of the leaf; Step 4: Convert the leaf to a PNG image; Step 5: Input the image into the computer vision tobacco leaf ripeness recognition model to obtain a feature image; Step 6: Obtain the main body of the leaf based on the feature image anchor box and leaf smearing area, and archive the recognized image to form a tobacco leaf archive; Step 7: Calculate the labeled area of different ripeness levels on both sides of the leaf, input the area size of different areas into the classification model to obtain tobacco classification; Step 8: When the tobacco classification meets the target classification requirement, the leaf is output.

5. The method of detecting tobacco leaves in a computer vision based pipeline automatic tobacco curing furnace according to claim 4, wherein, The step 2 further comprises: dividing the computer vision tobacco leaf recognition classification dataset into a training set and a test set according to an 8:2 ratio, and inputting the training set into a Kmeans model to train a classification model.

6. The method of claim 5, wherein the method further comprises: The computer vision tobacco leaf ripeness recognition model includes 11 layers, wherein: The first layer is a convolution layer Conv, which is a single convolution layer with an input of 640*640*3 matrix, an output channel number of 64, a convolution kernel size k of 3, and a stride step of 2; The second layer is a convolution layer Conv, which is a single convolution layer with an input of the output of the first layer, an output channel number of 128, a convolution kernel size k of 3, and a stride step of 2; The third layer is a channel to pixel layer C2f, which is three layers of C2f connected by three C2f modules, with an input of the output of the second layer, an output channel number of 128, and a Bottleneck shortcut; The fourth layer is a convolution layer Conv, which is a single convolution layer with an input of the output of the third layer, an output channel number of 256, a convolution kernel size k of 3, and a stride step of 2; The 5th layer is a channel-to-pixel layer C2f, which is a six-layer C2f layer connected by six C2f modules, takes the output of the 4th layer as input, outputs 256 channels, and the bottleneck has a shortcut; The 6th layer is a convolutional attention CBAM layer, which is a single-layer CBAM layer, has 256 channels, and a pooling kernel size of 7; The 7th layer is a convolutional layer Conv, which is a single-layer convolutional layer, takes the output of the 6th layer as input, outputs 512 channels, has a convolution kernel size k of 3, and a stride step of 2; The 8th layer is a channel-to-pixel layer C2f, which is a six-layer C2f layer connected by six C2f modules, takes the output of the 7th layer as input, outputs 512 channels, and the bottleneck has a shortcut; The 9th layer is a convolutional layer Conv, which is a single-layer convolutional layer, takes the output of the 8th layer as input, outputs 1024 channels, has a convolution kernel size k of 3, and a stride step of 2; The 10th layer is a channel-to-pixel layer C2f, which is a three-layer C2f layer connected by three C2f modules, takes the output of the 9th layer as input, outputs 1024 channels, and the bottleneck has a shortcut; The 11th layer is a spatial pyramid pooling layer SPPF, which takes the output of the 10th layer as input, has a pooling kernel size of 5, and outputs a feature map size of 20*20*1024.

Images